Further LWIP-related fixes

[dhalbert]

• Fixes Hard crashes on Pico W with high network activiity #10618

---

• Make all port_* storage operations use critical sections, not just those for LWIP.
• Fix a bug in espressif that was raising the wrong error code on a connect() failure. Discovered in passing.
• Incorporate some recent fixes from MicroPython extmod/modlwip.c.

This is a draft for now because there still seem to be occasional crashes on CYW43, and I want to understand why a connect failure is ECONNRESET instead of ECONNREFUSED, as it is on Linux.

@anecdata FYI

[anecdata]

Running Pico W against macOS server, looking good so far.

(Having issues updating a QT Py S3, will have to work on that one later.)

[dhalbert]

(Having issues updating a QT Py S3, will have to work on that one later.)

You may need to update the bootloader if it is a 4MB flash QT Py: https://learn.adafruit.com/adafruit-qt-py-esp32-s3/update-tinyuf2-bootloader-for-circuitpython-10-4mb-boards-only

[anecdata]

I did, with erase, several times (esptool and webserial), but the CP firmware doesn't take after that (it lights the neopixel solid orange and ejects all other USB devices). I'll try a different board.

update: working now with a reverse TFT S3. Both the PicoW and S3 are hitting the macOS server. Both clients are also still doing PulseOut just for fun. Looking good so far.

[anecdata]

Pico W and S3 have been running solid >16 hours, no hard faults, occasional client exceptions (mostly ETIMEDOUT, some I/O Error, and a few EINPROGRESS on Pico W; mostly ETIMEDOUT and a few EINPROGRESS on S3).

[dhalbert]

There was a CI failure on one board on the first run, unrelated to the PR. For some reason I was unable to re-run just that job: CI tried to re-run everything, which I cancelled.

[anecdata]

Looks good to me from an operational / testing perspective.

[tannewt]

Looks fine to me once the CI is happy. Maybe just push a new commit for testing?

[dhalbert]

The CI failure was just some one-off CI timeout.